Power: The Catastrophic Optical Damage
D I S S E R T A T I O N
zur Erlangung des akademischen Grades Dr. rer. nat.
im Fach Physik eingereicht an der
Mathematisch-Naturwissenschaftlichen Fakultät I Humboldt-Universität zu Berlin
von
Dipl.-Phys. Martin Hempel
Präsident der Humboldt-Universität zu Berlin:
Prof. Dr. Jan-Hendrik Olbertz
Dekan der Mathematisch-Naturwissenschaftlichen Fakultät I:
Prof. Stefan Hecht, PhD Gutachter:
1. Prof. Dr. Thomas Elsässer 2. Prof. Dr. William T. Masselink 3. Prof. Dr. Joachim Wagner eingereicht am: 07.05.2013
Tag der mündlichen Prüfung: 23. Oktober 2013
Zusammenfassung v
1 Introduction 1
2 Physical Background 3
2.1 Semiconductor Diode Lasers . . . 3
2.1.1 Development of the Structure of the Semiconductor Diode Laser . . . 3
2.1.2 Laser Mechanisms . . . 5
2.1.3 High-Power Diode Lasers . . . 8
2.2 The Catastrophic Optical Damage (COD) . . . 9
3 Experimental 13 3.1 Thermocamera Based Setups . . . 13
3.1.1 Thermocamera . . . 13
3.1.2 Specific Setups . . . 14
3.1.3 System Characterization . . . 16
3.1.4 Basic Image Processing . . . 20
3.2 Streak Camera Based Setups . . . 22
3.3 The Step Test Approach . . . 23
3.4 Further Analytical Techniques . . . 24
3.5 Samples . . . 26
4 First Phase of COD – Aging 29 4.1 Experimental Results . . . 30
4.1.1 Spatially Integrated Transient Data and Thermography . . . 30
4.1.2 Spatially and Temporally Resolved Data . . . 32
4.1.3 Active Layer Temperature Transient . . . 35
4.1.4 Influence of the Material System . . . 38
4.2 Discussion of First Phase Results . . . 40
4.2.1 The Time to COD . . . 40
4.2.2 Beam Propertiesen routeto COD . . . 42
4.2.3 Temperature Contributions . . . 43
4.2.4 Influence of the Material System . . . 45
4.2.5 Mechanisms Determining the First Phase . . . 47
5 Second Phase of COD – Thermal Runaway and Start of Defect Spread 49
5.1 Experimental Results . . . 49
5.1.1 Time Constants for Different Material Systems . . . 49
5.1.2 Spatial and Temporal Resolved Nearfield Dynamics . . . 51
5.1.3 Microscopic Study of Early Defect Stages . . . 60
5.1.4 Short-Time Thermography . . . 63
5.2 Discussion of Second Phase Results . . . 64
5.2.1 Spatially and Temporally Resolved Dynamics . . . 64
5.2.2 Energy Balance of COD . . . 68
5.2.3 Microscopic Defect Analysis . . . 76
5.2.4 Heat Dissipation . . . 81
5.2.5 Mechanisms Determining the Second Phase . . . 83
6 Third Phase of COD – Defect Spread 85 6.1 Experimental Results . . . 85
6.1.1 Defect Analysis Through the Top Contact . . . 85
6.1.2 Simultaneous Analysis of Both Facets and Cavity . . . 89
6.2 Discussion of Third Phase Results . . . 92
6.2.1 Data Obtained Through the Top-Contact Window . . . 92
6.2.2 Data Obtained From Both Facets and Cavity . . . 100
6.3 Modeling . . . 101
6.3.1 Introduction of the Model . . . 102
6.3.2 Influence of the Model Parameters . . . 105
6.3.3 Application of the Model . . . 107
6.3.4 Evaluation of the Model . . . 114
6.4 Mechanisms Determining the Third Phase . . . 114
7 Investigation of Special Structures 117 7.1 Laser Bars . . . 117
7.1.1 Experimental Results . . . 117
7.1.2 Discussion of the Results Obtained with Laser Bars . . . 118
7.2 Single-Spatial-Mode Ridge Waveguide Diode Lasers . . . 121
7.2.1 Experimental Results . . . 121
7.2.2 Discussion of the Ridge Waveguide Laser Dynamics . . . 125
7.2.3 Comparison to BA Diode Lasers . . . 128
7.3 Quantum Dot Devices . . . 129
8 Conclusions 133
List of Abbreviations and Symbols 135
Publications 139
Bibliography 143
Halbleiter-Diodenlaser spielen eine wichtige Rolle in allen Bereichen des täglichen Lebens, wie Da- tenspeicherung, Beleuchtung, Sicherheitstechnik, Markierungs-, Mess- und Drucktechnik. Sie stellen heutzutage die wichtigste Quelle für Laserstrahlung dar, entweder als direkte Strahlungsemitter oder als Pumplaser in Laser-Systemen. Hochleistungs-Diodenlaser sind die effizienteste menschgemachte Struk- tur zur Umsetzung von elektrischer in Strahlungs-Energie. Sie erreichen eine Effizienz von≤73% bei Raumtemperatur. Dieser hohe Wirkungsgrad führt nicht nur zu Energieeinsparungen, sondern reduziert gleichzeitig die Erwärmung des Bauelements durch Verminderung der Verluste.
Das Erreichen hoher optischer Ausgangsleistungen ermöglicht die Verwendung der Diodenlaser in einem breiten Anwendungsfeld. Eine herausragende Bedeutung dabei nimmt die Verwendung als Pump- Lichtquelle für Festkörper-Laser ein. Bei dieser indirekten Anwendung führt das Erzielen einer höheren Leistung pro Pumplaser-Emitter direkt zu einer Kosten- und Größenreduzierung des Gesamtsystems aufgrund der verringerten Anzahl benötigter Pumplaser. Beispiele für eine direkte Nutzung der erzeugten Strahlung mit hoher optischer Leistung sind die Verwendung als Seed-Laser, Materialbearbeitung, wie das Plastik-Schweißen, sowie medizinische und kosmetische Anwendungen.
Die Mechanismen, welche die maximal erreichbaren optischen Ausgangsleistung begrenzen, können in zwei Gruppen unterteilt werden, solche, welche die optische Leistung reversibel reduzieren und die- jenigen, die zu irreversiblen Degradation führen. Der catastrophic optical damage (COD) gehört zur letzteren Gruppe. Nachdem viele, die Bauelemente-Lebensdauer bei hohen optischen Leistungen be- grenzenden Faktoren, wie die Materialqualität, überwunden worden sind, bleibt der COD als ein funda- mentales Limit. Das Ziel der vorliegenden Arbeit besteht in der Analyse der mikroskopischen Ursachen des COD und des Verständnisses der damit einhergehenden physikalischen Gesetzmäßigkeiten. Dies soll zu wissensbasierten Lösungen zur Verbesserung der Bauelemente Effizienz, optischen Leistung und Zu- verlässigkeit führen. Damit ließen sich Anwendungen erschließen, die entweder auf niedrigen Preisen beruhen, z.B. ein Produkt für den Massenmarkt wie laserbasierte Zündkerzen, oder Systeme, welche ei- ne gewaltige Anzahl an Pump-Lasern benötigen, wie Einrichtungen zur lasergestützten Fusion (mehrere Millionen Laserarrays werden hierfür benötigt). Diese Einsatzfelder haben offenkundig das Potenzial tiefgreifender Auswirkungen auf die ganze Gesellschaft.
In dieser Arbeit wird der COD zeitaufgelöst untersucht um die beteiligten Mechanismen zu identifi- zieren. Zu diesem Zweck wird der COD mittels kurzen Pulsen bei hohen Strömen gezielt hervorgerufen.
Die Annäherung an die COD-Schwelle des jeweiligen Bauelements wird durch aufeinanderfolgende kur- ze Einzelpulse mit sukzessiv steigenden Strömen realisiert. Dieses Verfahren wird Stufentest genannt.
Der prinzipielle Vorteil dieses Verfahrens ist die Möglichkeit den Zeitpunkt des COD-Auftretens relativ genau vorherzubestimmen. Die Nutzung des Stufentests zielt dabei nicht nur auf die Erforschung des gepulsten Laserbetriebs ab. Es wird in der Arbeit gezeigt, dass der Stufentest eine besondere Form der beschleunigten Alterung darstellt und auf den gleichen Mechanismen beruht, welche auch im Dauer- strichbetrieb zum COD führen. Damit sind die erhaltenen Erkenntnisse auch für andere Betriebsregime relevant.
Der COD Prozess konnte zeitlich in drei Phasen unterteilt werden, die Alterung, der thermische Ru- naway (ein sich durch positive Rückkopplung selbst verstärkender Prozess) und das Sekundärschadens- wachstum. Die erste Phase konnte durch den Stufentest auf den Nanosekunden-Bereich reduziert werden.
Die Rolle des Laser-Lichtfelds als wichtigste Energiequelle des COD-Prozesses wurde experimentell be- stätigt. Die genutzten thermographischen Techniken erlaubten einein-situVerfolgung des Defektwachs- tums, auch wenn das Bauelement nicht, z.B. durch Öffnungen in den Metallkontakten, speziell dafür prä- pariert wurde. Diese direkte Messung der Ausbreitung, die Modellierung des Wärmeflusses und eine kris- tallographische Materialanalyse zeigen, dass das Material, welches von der Defekt-Front passiert wurde, innerhalb von Nanosekunden zu substanziell tieferen Temperaturen zurückkehrt. Verschiedene experi- mentelle Ansätze bestätigen das Vorhandensein einer Temperatur im Bereich von 1200◦C−1500◦C wäh- rend des gesamten Degradationsprozesses. Dabei hat sich gezeigt, dass selbst wenn keine Laseremission mehr vorliegt, die verstärkte spontane Emission ausreicht, um den fortschreitenden Degradationsprozess mit Energie zu versorgen.
Für den Start des thermischen Runaway muss ein bestimmter Temperaturunterschied zwischen der späteren COD-Position und dem übrigen aktiven Lasermaterial erreicht werden. Die vorliegende Arbeit zeigt verschiedene Mechanismen auf, die zu einer solchen Situation führen können. Dabei spielen auch physikalische Eigenschaften der verwendeten Materialsysteme und Schichtstrukturen eine entscheiden- de Rolle. Die Positionen dieser COD-Startpunkte haben entscheidenden Einfluss auf das spätere voll- entwickelte Schadensbild. Ein neu im Rahmen dieser Arbeit entwickeltes Modell zur Beschreibung der räumlichen Schadensausbreitung nutzt diesen Umstand, um die Defektkinetikex-postzu rekonstruieren.
Dies ermöglicht das Aufzeigen von Schwachstellen im Bauelement. Aus einer Analyse der Schadensaus- breitung während der zweiten und dritten COD-Phase konnte der thermische Widerstand des Materials als bestimmender Parameter für die Schadens-Dynamik identifiziert werden.
Experimentelle Befunde weisen darauf hin, dass der COD nicht am Quantengraben selbst beginnt, son- dern im Bereich des Wellenleiters und der Mantelschicht. Dies ist überraschend, da die höchste Lichtin- tensität am Quantengraben als Verstärkermedium zu erwarten ist.
Die vorliegende Arbeit stellt experimentelle Strategien zu Präparation sehr früher Defektstadien vor, welche zukünftig dazu genutzt werden können gezielt frühe Degradationsstufen zu untersuchen. Dabei sind vor allem Untersuchungen zur Alterung in Systemen mit externer optischer Rückkopplung ein in- teressantes Anwendungsfeld. Dies ist eine Fragestellung, welche aufgrund der heutzutage erzielbaren hohen optischen Leistungen für nahezu alle Diodenlaser-Systeme relevant ist, da eine Rückkopplung durch teilweise Strahlreflexion an Strahlkombinations- und Formungs-Optiken auftritt.
Semiconductor diode lasers are present in almost all fields of our daily life, e.g., data storage, marking, measuring, printing, lighting, security etc. They are the main source of laser light today, either as direct source or as pump source in laser systems. High-power diode lasers are the most efficient man-made structures for conversion of electrical to optical energy. Their efficiencies reach values of≥73% [1, 2]
at room temperature. This leads not only to energy savings, but reduces heating by avoiding losses.
Reaching high output power levels opens up a wide variety of application fields. An important example is the pumping of solid state lasers [3]. For this application, higher power from a pump laser emitter results in cost and size reduction of the entire system due to a reduced number of necessary pump lasers.
Examples for the direct utilization of diode lasers with high optical output power are the use as seed lasers, the treatment of materials as plastic welding, as well as medical and cosmetic applications.
The limiting factors for reaching higher output powers can be separated into two groups: Mechanisms that reduce the output power in a reversible way and irreversible degradation [4, 5]. The catastrophic optical damage (COD) belongs to the second group. After having sorted out many other factors, e.g., material quality, that impairs device reliability at high power levels, the COD remains as one of the main limitations. This thesis aims at revealing the microscopic origins of the COD and understanding its physics. This should lead to knowledge-based solutions for increasing the efficiency, output power, and reliability. This could enable applications that either rely on low prices, e.g., a mass-market product like laser driven spark plugs [6], or systems that need an enormous amount of pump lasers, as laser induced fusion facilities (several million laser arrays are required) [7]. Obviously, these options could affect the entire society.
The strategy followed in this thesis is to time-resolve the COD in order to understand the mechanisms leading to and being involved in the process. The COD is provoked with individual short high-current pulses. This addresses directly the pulsed operation regime, but also allows conclusions, as will be shown, about the COD failure in continuous wave operation representing the primary operation mode.
The main advantage of the single pulse test is the possibility to trigger the start of the COD event. This is important for all kinds of measurements with high temporal resolution.
The main benefits of the single pulse approach are a direct analysis of the defect growth kinetics as well as the temperature transients during COD. Moreover, it makes the preparation of extremely early defect stages possible. By stopping the degradation process at the end of the pulse, the pulse width determines the temporal resolution of the measurement but not the technique applied to analyze the defect. Furthermore, this approach allows to study the COD by analyzing the incorporated energies by the control over the external parameters.
This thesis is organized as follows: After an introduction of the semiconductor diode laser and a review of existing knowledge about the COD process in Ch. 2, the experimental techniques are presented in Ch. 3. The analysis of the mechanisms involved in COD follows its three phases which will be introduced at the beginning of Ch. 4. Chapter 4 is dedicated to the first phase, while Ch. 5 addressed the second, and Ch. 6 the third. The universality of the uncovered mechanisms is verified in Ch. 7, where special device structures are investigated. Chapter 8 gives conclusions and an outlook.
2.1 Semiconductor Diode Lasers
2.1.1 Development of the Structure of the Semiconductor Diode Laser
The first semiconductor diode lasers (DL) have been presented in 1962 [8–11]. These structures con- sisted of a pn-junction formed by zinc diffusion into n-type GaAs. As today’s edge-emitters, they had a Fabry-Perot-type laser cavity. These lasers suffered from two major problems [12]: Insufficient carrier confinement and unacceptable high optical losses. The transversal extent of the pn-junction is governed by the sum of minority carrier diffusion lengths. Additionally the light experiences high losses when propagating outside this gain region. The optical confinement in lateral direction was defined by the width of the entire chip. These problems result in a high threshold current (Ith) and make continuous wave (cw) operation of this structure impossible at room temperature.
An improved transversal carrier confinement is achieved by introducing energy barriers in the band structure [13]. In thisdouble heterostructureDL, a lower band gap material is surrounded by one with a wider band gap. Figure 2.1(a) shows a schematic of the conduction band edge for Al(GaAs)-material.
This approach also avoids high optical losses by absorption by using wider band gap material, which is transparent for the laser light. The use of AlxGa1−xAs material was favored by the almost matching lat- tice constants of AlAs and GaAs. This allows material growth with low defect densities due to an almost vanishing built-in strain. Room-temperature cw operation of DLs have been demonstrated employing this concept [14, 15].
GaAs GaAs GaAsAlGaAsx1-x AlGaAsx1-x
(a)
GaAs GaAs GaAsAlGaAsx1-x AlGaAsx1-x
AlGaAsy1-y AlGaAsy1-y
y<x (b)
GaAs GaAs GaAsAlGaAsx1-x AlGaAsx1-x
AlGaAsy1-y AlGaAsy1-y
y<x (c)
y
Figure 2.1: Schemes illustrating the various laser layer designs. The conduction band edge in transversal direction is given for: (a) Double heterostructure. (b) Separate confinement heterostructure.
(c) QW heterostructure. As example, possible compositions of the AlGaAs-system are given.
The idea for the graph is taken from [12, 16].
This idea was developed further in order to achieve separate control of carrier and optical confinement.
Additional layers with even lower refractive index were added on both sides of the double heterostructure.
These additional layers form an optical waveguide. The resulting structure is calledseparate confinement heterostructure[17]. Figure 2.1(b) schematically shows the conduction band edge for this case.
The use of molecular beam epitaxy and metalorganic chemical vapor deposition allowed for better
controllable and high quality growth of the DL structures. They allow to reduce the gain region thickness from≈50 nm to aquantum well(QW) of about 5−10 nm [18]. A scheme of the conduction band edge of this structure is given in Fig. 2.1(c). Structures with multiple QWs forming the gain region are used, too.
Control of the lateral dimension was achieved by restricting the current (I) path to a narrow stripe [19]. This principle is calledgain-guiding. The gain-guided lateral emitter width is influenced by the current spread under the contact stripe. The final width results from a gain-induced increase of the refractive index in lateral direction and is distorted by carrier-induced anti-guiding. The latter is a result of a decrease of the refractive index (n) in the material with increasing carrier density. This behavior is qualitatively discussed by Bennettet al. [20]; the authors identify the relevant mechanisms that affect nto be: (i) band-filling, resulting in the Burstein-Moss effect (detected as absorption decrease slightly above the band gap energy due to conduction band filling), (ii) band gap shrinkage which is an effect of the many-body interactions of electrons and holes at their band edges, and (iii) free-carrier absorption, which is, following the Drude theory, proportional to the density of electrons and holes and to the square of the wavelength.
These mechanisms, however, lead to an unstable lateral mode pattern in broad area (BA) emitters.
A solution for this issue was the application of an built-in lateral refractive index step in the structure.
Two methods became popular to achieve this: (i) the buried heterostructure laser, where after planar growth of the layers an emitter stripe is formed by completely removing the active layer outside and embedding the stripe thereafter in a material with lowern, and (ii) theridge waveguidelaser, where only the cladding layer outside the emitter stripe is thinned after growth. This results in an effective reduction of the refractive index in the region with thinned claddings due to compression of the optical mode in transversal direction there.
Other material systems have been used to build DLs for other wavelength than≈800 nm. InGaAsP double heterostructure lasers emitting at 650 nm have been realized by Colemanet al.[21] and AlGaInP around 670 nm, also as double heterostructure, by Ikeda et al. [22]. Additionally, Yablonovitchet al.
[23] and Adams [24] came up with the idea to use a strained QW by utilizing the lattice mismatch of the materials. Lattice mismatched materials allow a good, dislocation free growth as long as the strained layer does not exceed a critical layer thickness, as demonstrated by Matthews and Blakeslee [25]. In this growth regime, the lattice mismatched material accommodates to the host lattice constant and therefore builds up an intrinsic strain. This additional design freedom allows to reach the wavelength range from 980 nm to 1100 nm, e.g., demonstrated by Beernink et al. [26]. The blue and ultraviolet wavelength ranges have been accessed by the invention of DLs based on GaN by Nakamuraet al. [27]. Wavelength of 1.3µm and the important wavelength of 1.55µm for fibre communication systems are realized in GaInAs on InP [28].
By further development and understanding of the growth mechanisms, other structures of the active region became possible. An approach, presented by Arakawaet al.[29], is to use quantum dots (QD) as the active material. Another approach is the quantum cascade laser (QCL) using intra-band transitions for lasing. A first successful experimental realization of it was reported by Faistet al.[30] in 1994. They realized coherent light amplification of the radiative relaxation of electrons from a higher to a lower state within the conduction band.
Typically, a fraction of≈25%−50% of the electrical power applied to a DL is converted into heat.
Therefore, cooling of the active region is important. Especially BA lasers are packaged with their epi-
taxial layers close to the heat sink (also denoted as sub-mount). Due to the common use of n-doped substrates, the p-side of the diode is on top of the grown layer stack. If now placing the epitaxial layers as close as possible to the heat sink, the p-side is directly attached to it. Therefore this kind of mounting is called p-down(the vice versa situation is called p-up mounting). In case of p-down mounting, it is common to package the laser chip with an overhang of some 10µm with respect to the heat sink. This is done to protect the front facet from solder material. Furthermore, the soldering techniques provide challenges, in particular for larger devices where a mismatch of the thermal expansion coefficients is crucial [31].
A photograph of a mounted DL is given in Fig. 2.2(a). There the GaAs/AlGaAs laser chip is mounted p-side down on a copper sub-mount. The front facet is directed towards the reader. The n-side metal- lization is connected by bond wires to a bond pad which is electrically insulated from the heat sink. This geometry is shown schematically in Fig. 2.2(b). The coordinate system used in the following chapters is also given – growth (transversal) direction: y, lateral direction: x, longitudinal direction/resonator axis:
z. The layer scheme of the separate confinement heterostructure is shown on an enlarged scale on the right side. The epitaxial layers have a thickness of.4µm in total, while the entire chip has a height of
≈120µm−160µm.
Substrate n-Cladding
p-Cladding Waveguide Waveguide QW
Bond Pad Submount/Heatsink Bond Wires
- +
(a) (b)
Figure 2.2: DL on sub-mount. (a) Photograph of a p-down packaged DL on a copper sub-mount. The bottom contact (p-side) is electrically connected by the sub-mount, while the top-(n)-contact is connected by bond wires to a bond pad. The front facet is directed towards the reader and appears dark. (b) Scheme of the p-down packaged laser (parts are labeled accordingly).
The blue arrow indicates the direction of light emission. The layer scheme of the separate confinement heterostructure is given at the right side. Their position is indicated in the layer scheme by a dotted rectangle.
2.1.2 Laser Mechanisms
The majority of today’s edge emitting semiconductor DLs are based on QWs as a gain medium. There- fore, it is worthwhile to discuss the physical properties of those structures with reduced dimensionality.
The relevant case for semiconductor DLs is the so-calledtype-I QW, where a lower (confining) potential in the QW is experienced by both electrons and holes. For achieving a reduction in dimensionality from three dimensions in the bulk semiconductor to a two dimensional confined state in a QW, its thickness
must be on the order of the de Broglie wavelengthλB=h/p
2m?cE(withh: Planck’s constant,m?c: ef- fective electron mass in the semiconductor; E: electron energy). This criterion is fulfilled by 5 nm to 10 nm thick QWs in DLs. The electrons in a QW can be treated as two-dimensional electron gas. In the envelope function approximation, the wave function for the electron in the QW can be separated into two contributions [32–34]:
Ψ(x,y,z) =ζ(y) ei(kxx+kzz)u(k=0,x,y,z) (2.1.1) with the lattice-periodic Bloch function u and the wave vector k and its components kxx andkzz. In the x−z-plane the particle exhibits free dispersion as in the bulk case. The part of the wave function along growth directionζ(y)equals the particle-in-a-box-problem forkxx=kzz=0, and the energy states become quantized. With this wave functionansatzthe possible energies of the system can be calculated [32]:
Ec,n=Ec0+Ey,n+h¯2 k2x+k2z 2m?c
infinite high barriers
=⇒ =Ec0+ h¯2
2m?c
nπ L
2
+k2x+k2z
, (2.1.2) with the conduction band edge Ec0, the quantum-number n=1,2,3, . . . indicating the sub-band, the quantized energy iny-directionEy,n, and the thicknessLof the QW. The quantum confinement increases the ground state energy by∆E= (¯h2/(2m?c))·(π2/L2)compared to the bulk case, because the momentum is increased by a decrease in spatial uncertainty following the uncertainty relation. Moreover, the excited state energies become quantized. A similar relation holds for the holes in the valence band.
The reduction of dimensionality by the quantum confinement also impacts the density of states of the system by reducing the available space of possible wave vectors [35]. The densities of states (ρc) for the conduction band of bulk material, QWs, and QDs are (with the lowest energyEc,nin sub-bandn):
bulk : ρc3D(E) =
√ 2m?c3/2 π2h¯3
√ E−Ec0 QW : ρc2D(E) = m?c
π h¯2L for Ec,n|k=0≤ E ≤Ec,(n+1)
k=0 else ρc2D(E) =0 QD : ρc0D(E) =δ(E−Ec,n)
(2.1.3)
Figure 2.3: Density of states for bulk material (3D), QWs (2D), and QDs (0D).
Figure 2.3 depicts the density of states for bulk material (3D), QWs (2D), and QDs (0D) [33]. While for the bulk materialρc3Dgives a square root starting fromEc0, it is step like for the QWs with the bulk ρc3Das limit. In case of the QDsρc0Dis a Dirac-Delta function at the sub-band energies. Similar relations holds for the valence band density of states ρv. They can be combined to the joint density of states ρj−1=ρc−1+ρv−1. This has been calculated for a single quasiparticle, an inclusion of Coulomb mediated many-body effects modifies the density of states, in particular at small electron energies [36].
When studying laser transitions only the first sub-bands in both conduction (Ec,1) and valence band (Ev,1) are of interest, caused by selection rules and occupation probability. The laser light frequencyν is determined by the transition energy between them, i.e., hν =Ec,1−Ev,1 which is the band gap Eg plus the confinement energies of the carriers. A quasi thermal equilibrium in the conduction and valence bands can be assumed, caused by the fast intraband thermalization (≈100 fs) compared to the interband relaxation (≈1 ns) [37]. Therefore, Fermi distributions are used to express the occupation probabilities in the conduction band (fc) and valance band (fv) [33]:
fc= 1
expE−E
f,c
kBT
+1
and fv= 1
expE
f,v−E kBT
+1
(2.1.4)
with the quasi Fermi energies for electrons Ef,c and holes Ef,v, the Boltzmann constant kB, and the TemperatureT. The gain in the laser is [37]:
g(hν)∝
|M(hν)|2
hν ρj(hν) (fc−fv) (2.1.5) The transition matrix element|M(hν)|2 is almost constant for all III-V-compounds [38]. The quantum confinement affects the gain byρj. The Fermi inversion factor finv= fc−fv depends on the electrical pumping. It gives a sufficient condition for achieving positive optical gain:
from finv>! 0 follows Ef,c−Ef,v>Ec,1−Ev,1=hν (2.1.6) A second lasing condition arises from the losses in the Fabry-Perot cavity. The modal gainΓg has to compensate at least the internal (absorption) lossesαi and the mirror lossesαm:
Γg≥αi+αm=αi+ 1 2Lcavln
1 R1R2
(2.1.7) with the confinement factorΓwhich is the ratio between the volume of the gain medium and the volume of the optical mode (typically≈1% for DLs). Further parameters are the cavity lengthLcavand the front and rear facet reflectivitiesR1andR2, respectively.
The impact ofρjon the gain spectrum is shown in Fig. 2.4. The joint density of states for bulk material and QW are schematically shown in Fig. 2.4(a). The Fermi inversion factor on the same energy scale is given in Fig. 2.4(b), the zero-crossing of this curve is represents the sufficient lasing condition discussed above. Figure 2.4(c) gives a comparison of the gain curves for the bulk and QW case. There shapes are determined by the product ofρjand finv, as demonstrated in Eq. (2.1.5). The gain spectrum of the QW is narrower than in the bulk case and the peak gain is lower. In order to satisfy the second condition for lasing, the compensation of the lossesαi+αmthe pumping has to be increased.
hn
hn
hn Bulk
QW
Ef,c-Ef,v +1
-1
LossGain
0 g h( n) finv(hn)
QW Bulk rj(hn)
a ai+ m
g h( n) g h( n)
hn
hn Bulk
QW (a)
(b)
(c)
(d)
(e) 0
0 0
0 Eg
E>Eg Eg
N
N
Figure 2.4: Comparison of gain spectra for bulk material and QW lasers. (a) Joint density of states. (b) Fermi inversion factor. (c) Gain. (d) Evolution of bulk material gain curve with varying carrier injection. (e) Gain curves of the QW with varying carrier injection. The idea for (a-c) is taken from [33].
Figures 2.4(d,e) show the evolution of the gain curves for bulk material and QW lasers with variation of the carrier injectionN. The QW lasers satisfy the lasing threshold condition earlier, caused by the reduced density of states that has to be filled.
In QD lasers the lasing conditions are expected to be satisfied at even lower injection densities, because of the further reduced number of states to be populated. This is true, as long as the gain material contains only QDs of one size which have the same confinement energies. Otherwise, the Dirac-Delta peaks shown in Fig. 2.3 are distributed over a wider energy range negating the advantages of the strong quantum confinement.
2.1.3 High-Power Diode Lasers
In many fields of application, the optical output power of the DLs is a key parameter. Therefore, there is a permanent demand for further increasing emission powers. The internal power density of state-of-the-art devices amounts up to 1010W/cm3, which is higher than in the core region of the sun. Every success in pushing the extreme limits of technological feasibility further has an direct impact, e.g., on the size and
efficiency of laser systems that use DLs as pump source. Two standard applications in this field are the pumping of Nd:YAG at 808 nm and erbium-doped fiber amplifiers at 980 nm. For this reason, devices of these wavelengths cover the main part of the experimental work presented in the following.
Today’s DLs are the most efficient man-made structures for converting electrical power to light with values exceeding 73% efficiency [1, 2]. Nevertheless, the remaining supplied electrical power remains in the device, leading to a heating of the semiconductor structure. Therefore, the achievable upper limits of high-power DLs operation are closely connected to cooling concepts. A common strategy is the p- down mounting to allow an efficient heat flow from the active region to the heat sink. The latter can be passively cooled, i.e., via heat conduction, but also actively cooled via a micro-channel water cooler.
Another difference compared to low-power DLs is the gain saturation at high current injection, which will be discussed in Sec. 3.5. Moreover, a way to extract more power from a single device is to increase the emitter width. BA DLs are a result of this powerful concept. The main drawback from this approach is the reduced beam quality by allowing the propagation of several lateral modes in the wide emitters.
This leads to instabilities in the beam, known asfilamentation. The characterization and control of it is still subject to on-going work. However, the maximal width of these structure has also an upper limit to allow sufficient beam guiding. In order, to achieve an higher output power in the kW-range from one monolithic device, arrays are formed by integrating several single emitters on one single chip. The most widespread concept is the parallel alignment of the emitter stripes, but also coupled waveguide structures have been tested.
Beside the decreased beam quality, high-power DLs face problems with the high internal optical load in the device. It is reduced by application of an efficient anti-reflective coating on the out-coupling facet.
In order to maintain at the same time the high efficiency and output power level of the device, the cavity length is increased to compensate the elevated mirror losses and the epitaxial structure is optimized for low internal optical losses [39].
2.2 The Catastrophic Optical Damage (COD)
Failure mechanisms observed in DLs have been categorized by the time-evolution of the output power loss during operation with constant current, see Fig. 2.5(a). This leads to three main groups, as discussed by Jimenéz [40]:
(i) Rapid degradation is characterized by a quick decrease of the optical output power. This failure mode indicates significant defects, e.g., cracks or large dislocation networks, already present after device manufacturing.
(ii) Gradualdegradation results in a slow output power decrease. This is a characteristic behavior for regular operation. It includes point defect creation and defect motion through the material, both on long time scales.
(iii) Sudden degradation, where the device shows regular operation and then experiences an almost instantaneous power drop. This kind of degradation gets activated when reaching a threshold condition that triggers the start of a fast degradation process.
The COD belongs to the third group. It is a sudden failure mechanism of edge-emitting DLs and occurs at elevated emission power levels. Together with a substantial power loss, structural damage is detectable.
10 µm
Outputpower Time
I=const.
i
ii iii
(a) (b)
(c) (d) (e)
pristineCOD
Figure 2.5: (a) Three different degradation mechanisms, categorized by the output power evolution dur- ing operation with constant current: (i) Rapid degradation, (ii) Gradual degradation, (iii) Sudden degradation. (b,c) The front facet of a pristine 808 nm emitting DL (device A8) is shown. While (b) gives the lasing light emission, (c) is a micro-photograph of the surface.
(d,e) The front facet is shown after COD occurred at the same device as in (b,c). (d) The laser light emission with COD damage taken under same conditions as (b). (e) Micro-photograph of the front facet impaired by COD.
An example is shown in Figs. 2.5(b-e). Figures 2.5(b,c) gives the situation for the pristine device – the emission pattern of the front facet (b) and the microscope image of the facet (c). In Figs. 2.5(d,e), the same is given for the device impaired by COD. The loss of optical power is clearly visible as dark areas in the emission pattern (d). Furthermore, the damage at the facet is obvious in Fig. 2.5(e). The main driving force of the COD is the optical light field, as shown by Chinet al. [41] and Bou Sanayeh [42]. They demonstrated that the defect growth during COD follows the optical mode direction and not, e.g., a crystal axis. Therefore, the COD is an optically driven degradation process and should not be confused with other types of sudden degradation mechanisms such as electrical failures like the electrostatic discharge etc.
The emission power at which COD starts is commonly called COD threshold power (PCOD). The optical output power is related to the currentI via the DL characteristics. Therefore, a COD threshold current (ICOD) can be given. In this way, the termCOD thresholdis for bothPCOD andICOD.
Short after the first realization of the DL, reports were published addressing the observation of COD by Cooper et al. [43] in 1966 and Kresselet al. [44] in 1967. In 1973 Eliseev [45] characterized it as thermal micro-explosion. A comprehensive work, also addressing the origins of COD and its kinetics, has been published by Henry et al. in 1979 [46]. The main findings of this study are still valid for recent DLs: COD is jump-started by a fast thermal runaway. This process is initialized by reaching a certain temperature at the later COD site, named critical temperature (Tcrit). The first direct experimental verification of this has been reported by Tang et al. [47]. In later studies [48–52], the influence of extrinsic effects was shown such as surface recombination and creation of structural defects.
The precondition for COD is an elevated local temperature. This can be caused by different heating mechanisms. At the facets, the dominating ones are surface recombination (as demonstrated by Ziegleret al.[53]) and surface currents (Tanget al.[54]). Moreover, the packaging with overhang, as discussed in Sec. 2.1.1, results in an elevated front facet temperature in comparison to the bulk. This is due to the lack of a good thermal coupling to the heat sink in this area. Yooet al.[55] have found, that this temperature distribution is rather the ’normal’ case. Furthermore, it explains why the facets are commonly affected by COD. Nevertheless, defects in the interior of the cavity can also lead to a local temperature rise resulting in a thermal runaway.
Higher operation currents or aging induced defect creation in long term operation1 lead to a further increase of the temperature deviation between the later COD site and the remaining bulk material. This might be enough to reach Tcrit and to start the thermal runaway by closing a positive feedback loop.
This is characterized by a fast rise of the local temperature, which can be detected as a flash of Planck’s radiation as shown by Ziegleret al.[56]. The feedback loops can be categorized asintrinsicorextrinsic, even combinations of both are feasible [57, 58]:
The intrinsic loop, as described by Henryet al. [46], works as follows: The elevated temperature causes a local shift of the band gap. Therefore, the re-absorption of laser light is increased in this region. The result is a higher density of non-equilibrium carriers. This leads to a further temperature increase by non-radiative recombination of these electron-hole-pairs, even if the recombination rate keeps constant. An elevated temperature increases the re-absorption further, as will be discussed in Sec. 4.2.3, and therefore closes the feedback loop. In the case of QW lasers, this remains still an issue, although the modal absorption is about two orders of magnitude smaller than for the double heterostructure devices (Tanget al.[59]). As shown by Chenet al.[49], the barrier and waveguide materials of the QW DLs are also involved in the process. Furthermore, free carrier absorption is an additional mechanism supporting this feedback loop.
The extrinsic feedback loop relies on the fact that an elevated temperature stimulates defect accumu- lation and creation. This leads to an absorption of laser light via defect-related optical transitions or to absorption by macroscopic defects. Non-radiative recombination transfers the energy from the light field to the semiconductor lattice. This acts as additional heat source making further defect creation in- creasingly likely. The relevance of this kind of feedback is supported by reports about a lowered COD threshold in aged DLs [60–63]. Another evidence for this is theCOD re-ignitionby subsequent current pulses, that will be discussed in detail in Secs. 5.1.2 and 6.1.1.
Much effort was invested to prevent degradation during high power operation. The facets were identi- fied soon as one substantial bottleneck of the device structure. A major improvement was the application of facet coatings, allowing to adjust the reflectivities of front and rear facets (for as-cleaved facets it is about 30% determined by the refractive index change between semiconductor and air). Typically the rear is coated highly reflective (≈99% reflectivity), whereas the front facet has an anti-reflection coating (≈1−5% reflectivity). Common coating materials are dielectrics as Al2O3, applied in single or multi- layers onto the facets, a comprehensive description is given by Macleod [64]. The mechanism that makes the as-cleaved facets susceptible to COD was found to be a temperature-activated surface-chemical re- action, as reported by Latta and Moseret al.[65–67].
In order to shiftPCOD to even higher values, additional passivation of the as-cleaved facets became important. An example is theE2-technology, involving in-vacuum cleaving of the facets and low-energy deposition of silicon [68]. Approaches following a similar strategy use the deposition of germanium, aluminum, or antimony [69, 70] instead. An alternative approach is the cleaving of the facets in air and a subsequent cleaning, e.g., by ion-beam [71] or hydrogen-plasma [53] etching. This is followed by a passivation via nitridation [72] or sulphation [73–75]. Moreover, processes as InGaP or ZnSe epitaxy are used as well [76, 77].
1This results in an increased non-radiative recombination rate.
Beside the improvement of facet stability, a variety of additional methods have been developed to increase COD threshold:
• Low optical confinementstructures: In this devices, the stripe width is widened in vicinity of the facet [78, 79]. This reduces the optical power density there.
• The large optical cavity concept: The transversal confinement, i.e., the waveguide thickness, is widened [80, 81]. The aim of this approach is the same – reduction of the optical power density at the facet.
• The use of alternative material systems help reaching higher PCOD [82, 83]: The sensitivity to COD varies between them, so that a proper choice, if possible at the desired wavelength, helps to increase COD threshold.
• The application ofnon-injecting mirrorswhere the current density is reduced at the facet [51, 84–
87]: This reduces heating, e.g., by surface recombination. This structure is achieved by packaging the chip with overhang, cf. Sec. 2.1.1. Moreover, insulating layers can be applied close to the facet blocking the current flow there.
• In thenon-absorbing mirrorapproach, the aim is a reduction of the modal absorption in vicinity of the facet [78, 88–95]. Therefore, two common techniques are applied to QW lasers: (i) Ion implantation or a thermal treatment is used to shift the laser transition to higher energies. (ii) A tensile strained QW is used. This strain relaxes at the facet and results in a blue shift of the QW absorption.
The result of such efforts is the state-of-the-art cw PCOD value per aperture width of 285 mW/µm as reported by Petrescu-Prahovaet al.[81]. This characterizes COD as a generic mechanism related to high power densities. After facet technology reached such a high level, COD starting at other locations, e.g., in the bulk, became more relevant.
Nevertheless, there is a lack of microscopic understanding of the mechanisms involved in COD. There- fore, the approaches to avoid COD are partly developed by trial-and-error. In order to achieve a deeper understanding of the physics on a microscopic level, the work presented in this thesis has been under- taken.
In this chapter, the experimental techniques are presented that have been employed and in part introduced to study the COD. In the field of instrumentation, thermocamera- and streak camera-based setups are discussed in detail. Moreover, an overview is given of the in-house characterization techniques and of the external equipment.
3.1 Thermocamera Based Setups
The fact that COD is accompanied by a flash of Planck’s radiation (often referred to asthermal flash) favors the use of a thermocamera. This has been demonstrated by Ziegleret al.[56]. Therefore a couple of experiments were designed incorporating such a camera into the measurement setup.
3.1.1 Thermocamera
A major advantage of using a thermocamera, compared to a single point detector, is the additional geo- metric information and the possibility to monitor different parts of the device in parallel. In the experi- ments, a Thermosensorik CMT384 thermocamera was used. It is based on a HgCdTe focal plane array with 384 pixel×288 pixel of 20µm size in square, with a pixel-pitch of 24µm. The detector plane is cooled down to 80 K by a Stirling-cooler. The integration times can be chosen between 10µs and 5.1 ms in steps of 10µs. The noise equivalent temperature difference (NETD) is 20 mK. This is the temperature difference necessary to reach a signal-to-noise-ratio of one. The value given by the manufacturer is valid for a constant signal recorded with an integration time of 1 ms.
Two different objectives were used in the experiments. Thestandard one with a working distance of 0.1 m to infinity and themicroscope objective with 2.5× magnification and a working distance of
≈20 mm. In order to get the highest geometric resolution, the latter one was used. The resolution obtained with the microscope objective is 8.8µm×8.8µm per pixel. Notice, that the resolution in thermography is mainly limited by the longer light wavelength, e.g., in comparison to microscopy in the visible range. Therefore, a magnification of 2.5×, in conjunction with the given detectors’ pixel-size, is close to the theoretical resolution limit. The depth of focus while monitoring a DL is≈40µm, as will be shown in Sec. 6.1.2.
A second thermocamera was used for some of the experiments in collaboration with the Technical University of Denmark. The CEDIP Titanium 560M is based on a InSb focal plane array. It has 640 pixel×512 pixel with a pixel-pitch of 15µm. In conjunction with the 2.5×microscope objective this gives a resolution of 4.7µm per pixel. The spectral sensitivity is in the range from 3.6µm to 4.9µm.
The integration times are between 3µs and 20 ms, and the NETD is 20 mK. The use of this camera was analogous to the Thermosensorik CMT384. Results obtained with the CEDIP camera will be indicated, otherwise the Thermosensorik camera was used.
3.1.2 Specific Setups
Figures 3.1 and 3.2 show an overview of the thermocamera-based setups. A frame synchronous output signal is given by the thermocamera, cf. Fig. 3.3. It triggered a Stanford Research DG535 delay generator which delivered a delayed trigger signal to the DL driver PicoLas LDP-V50-100 V3, Fig. 3.2. The DL was directly attached to the driver in order to guarantee fast current rise times of ≤10 ns. Excited by the current pulse, the DL emited laser radiation and thermal radiation, Fig. 3.2. A dichroic mirror, being transparent for thermal radiation and highly reflective for laser radiation, was used under 45◦angle to separate this two signals. The laser emission was directed to a fast photo-diode (PD, ThorLABS DET10A/M, ≈1 ns rise time); neutral-density filters (ND) were used to damp the signal if necessary.
The PD signal was recorded by a fast oscilloscope (Agilent Infiniium, 2 GHz). The thermal radiation was detected by the thermocamera. Additionally, the current passing the diode was monitored and also recorded with the oscilloscope.
For mounting the DL, different configurations were used, as shown in Fig. 3.2(b-d). Figure 3.2(b) shows the typical situation for monitoring the front facet, the location where COD is most likely to start.
The laser emission was pointed directly towards the thermocamera. The cooled background shield gives an uniform thermal background which is desirable in thermography. In sub-figure 3.2(c) a configuration is shown that allows for inspection of front facet and top side at the same time, e.g., if the device has a window in the top-contact materialization. In order to image front and rear facet and additionally the side of the laser cavity at once, an arrangement as shown in Fig. 3.2(d) was chosen. Here, the DL was placed in a 90◦gold-coated corner-mirror.
If no spatial information is needed, the thermocamera can be replaced by a single-point MCT detector.
In this case a configuration as in Fig. 3.1 was used. The laser emission and near-infrared (NIR)-signal coming from the front facet of the laser were directed via an off-axis parabolic mirror onto point de- tectors. Both contributions were separated by a dichroic beam-splitter. The emission power transient was monitored by the fast PD (ThorLABS DET10A/M). The NIR emission was detected by the fast infrared detector Judson J15D22-M204-S250U-60 that ensures together with a fast preamplifier a time resolution of better than 100 ns. Filters formed a bandpass that allows detecting the 1.5−2.5µm NIR spectral range. The thermocamera was used in parallel to analyze the thermal radiation along the cavity by looking at the side of the laser stripe.
NIR PD
off-axis parabolic mirror
thermo-
camera DL
splitterbeam
Figure 3.1: Scheme of setup with thermocamera for inspection of the side of the cavity and two point detectors analyzing the radiation coming from the front facet in different spectral ranges.
mirror
submount
DL mirror
mirror submount
cooled background shield
DL DL
side view: side view: top view:
(b) (c) (d)
time integration
window
laser trigger (delayed) oscilloscope diode current monitor
photodiode
diode laser driver dichroic
mirror
delay generator
optics + ND filter microscope
objective
thermal image thermocamera
diode laser (a)
trigger
Figure 3.2: Thermocamera-based setups. (a) Overview of a typical experimental arrangement. The trig- ger hierarchy is indicated. The area around the DL is marked by a dotted line, here different configurations as shown in (b-d) are placed. (b) DL mounting for front facet inspection; green arrow indicates direction of laser emission. (c) Configuration that allows for measuring the top side of DL chip and the front facet simultaneously. (d) Placing the DL into 90◦ corner mirror allows imaging of front facet, rear facet and the side of the cavity in parallel; red arrows indicate thermal emission.
3.1.3 System Characterization
For use as thermal detector in COD experiments, the full sensitivity range of 1.5µm − 6µm of the thermocamera was limited by a filter to the 3.4µm − 5.5µm wavelength range. This prevented the recording of the GaAs-substrate-related emission below 2µm that has been characterized by Ziegleret al.[96].
Trigger Scheme
The camera is equipped with single-pulse trigger capability. It provides a camera frame synchronous output trigger signal, as schematically shown in Fig. 3.3(a). Due to the fact that neither the stability of this trigger nor its relation to the camera integration window were specified by the manufacturer, an additionally characterization was done. The details will be given in the following subsection. The frame synchronous output, provided by the thermocamera, is a square wave signal with negative amplitude, see Fig. 3.3(a). Its negative slope is used as time reference and set tot=0, see dashed line in Fig. 3.3.
The image integration timetint, i.e., the time window in which the camera-sensor is sensitive to imping- ing photons, starts with a delay that is determined to be 129.04µs, see Fig. 3.3(b). Details about the measurement of this delay will be given later. The delay generator DG535 is triggered by the signal at t=0. It provides two trigger pulses, one as trigger for the laser pulse generator (c), and another one as reference signal (d) att=0 in order to trigger further devices, e.g., the oscilloscope. With the help of these trigger-logic-signals, Fig. 3.3(a-d), the pulse generator is triggered and provides a single current pulse (e) to the laser diode. The pulse widthtPWis defined by the length of this current pulse. The delay is chosen in a way that the laser pulse is centered in the image integration window of the thermocamera.
It would be desirable, of cause, iftintandtPWmatch in length and temporal position. But, due to the fact, that the lowest given integration time of the thermocamera is 10µs and the preferred1 pulse length for the used single pulses is around 1µs, this alignment provided best results.
The output power was recorded by the PD. A typical result in case of COD is shown in Fig. 3.3(f).
The time where COD occurs, i.e., the significant output power drop, is marked by a star and labeled accordingly. The time interval from the leading edge of the pulse to the point where the power drop starts, i.e., the COD ignition, is denoted time to COD(tCOD), see Fig. 3.3(f). The time-resolved temperature profile of the COD site, in conjunction with the scenario shown in (f), is given in Fig. 3.3(g). The shape of the profile will be discussed in Ch. 4. The expected temperature transient is faster as the time resolution of the thermocamera. The recorded thermal signal is an average. If we consider only the time in which an elevated temperature is present an effective integration timetint,effcan be given, see indicated interval in Fig. 3.3(g).
Image Integration Window of the Thermocamera
In order to determine the camera trigger-timing and the sensitivity during a single pulse experiment, a fast-switchable mid-infrared light-source was used, namely a quantum cascade laser (QCL). It was emitting around 4.1µm and provided, together with the PicoLas LDP-V 50-100 V3 pulse driver, pulses of(18 ± 1)ns width atI=5.6 A. Details about the device are given in Ref. [34]. This short pulse was shifted in time by the delay generator with respect to the zero time mark, cf. Fig. 3.3.
1Details will be given in Ch. 4.
Frame Sync.
Output Trigger Image Integration Laser Trigger
Reference Trigger
Optical output power
Thermal radiation
Thermo- cameraDelay generator DG535
Diodelaser output
Time P
T tPW
tint
(a) (b)
(c) (d)
(f)
(g) LogicResponse
t = 0 129.04 µs
tint,eff
COD (e) I
Pulse generator Current Pulse
tCOD
Figure 3.3: Thermocamera setup – basic timing diagram. The trigger signals (a-d) are given, labeled as
’Logic’. The ordinate represents a logic level, e.g., in case of (c,d) TTL signals. The frame synchronous output trigger (a) is provided from the thermocamera. It has a fixed relation to the leading edge of the image integration window (tint) (b). The delay generator DG535 is triggered by the negative slope of (a) and provides a delayed laser trigger (c) and a delay- free reference trigger (d). The response to the trigger signal is shown in (e-g). (e) The current pulse delivered by the pulse generator; the ordinate gives the current. The length of the pulse defines the pulse width (tPW). (f) The optical power output (ordinate) in case of COD occurrence (marked by the star; after timetCOD). (g) Temporal evolution of the thermal radiation; the ordinate gives the temperature at the COD site. The effective integration time (tint,eff) starts with the leading edge of the current pulse. The indicated time intervals will be discussed in the text. All graphs are aligned in time, as indicated at the bottom. The corresponding devices are given at the right side.
For every delay interval, the single pulse emission was recorded with the thermocamera and integrated over all pixels. Due to the short duration of the laser pulse, compared to the integration window, a mathematical deconvolution of the data sets was proven to be unnecessary.
First, it was checked if the nominal integration times are in agreement with the real ones. The result for four different values is given in Fig. 3.4. Beside the nominal integration times of tint,nom =10µs, 40µs, and 100µs, the camera manufacturer provides a setting called ’0µs’. Its meaning becomes clear in Fig. 3.4. The integration time can be set in steps of 10µs, representing the interval between the leading and trailing edge of the integration window. In case oftint,nom=’0µs’, the window consists only of leading and trailing edge without any delay in between.
120 140 160 180 200 220 10
100 1000
'0 µs'
10 µs
40 µs
100 µs
Sumofthermalsignal(counts)
Time (µs)
10 100 1000 10
100 1000
tint
(µs)
t int,nom
(µs)
Figure 3.4: Thermocamera integration windows. Sum of thermal signal over time. The time reference is the negative slope of the signal shown in Fig. 3.3(a). The inset shows the relation between nominal integration timetint,nomand measured integration timetint.
By taking data for different integration times, the relation between the set valuetint,nom and the real windowtintwas verified to be linear, see inset in Fig. 3.4.
It is obvious, that there is a deviation from a rectangular shape of the camera integration window around its leading edge, see Fig. 3.4 (≈115µs−130µs). This deviation is negligible for integration times of≥10µs. However, it is worth to take a closer look at this interval for two reasons:
• to check the temporal stability of the position of the beginning of the integration window, and
• to characterize the camera integration in case oftint,nom=’0µs’.
Figure 3.5 shows the leading edge of the integration window on an expanded scale. In order to check the trigger stability not only for differenttint,nomvalues, but also for independent runs, the measurements for tint,nom=’0’µs and 10µs were repeated on another day. The curves of Fig. 3.5 are labeled with ’day 1’
and ’day 2’ (the remaining curves are taken on day 1).
The result is, that the leading edge of the integration window2is always at(129.040±0.008)µs. The jitter of the signal of 8 ns is thereby mainly caused by the laser current pulse generator. Moreover, the full width at half maximum (FWHM) of the ’0’µs integration window has a value of 130 ns. Here, the leading tail of the curve (<129.04µs) is relevant in contrast to the longer integration times. For practical measurements, a more realistic value of 460 ns is derived, considering the areas under the curve.
The problem with the use of such short integration times is the small number of photons contributing to the signal. As long as the NETD is dominated by thermal and shot noise, it scales in accordance with the integration time. Following this scaling, Tab. 3.1 gives an impression of the NETD-values (based on the given value fortint=1 ms). However, these values are valid for the average temperature of the area that is detected by one camera pixel and has to be scaled accordingly if the heat source becomes smaller.
2It is the time, where the signal reaches the half of the value of the constant plateau that starts at≈129.3µs.
128.6 128.7 128.8 128.9 129.0 129.1 129.2 129.3 129.4 100
1000
Sumofthermalsignal(counts)
Delay time (µs) '0 µs', day 1
'0 µs', day 2
10 µs, day 1
10 µs, day 2
40 µs
100 µs
1000 µs
Start at 129.040 µs
FWHM ('0 µs'):
130 ns
Figure 3.5: Leading pulse edge of data as shown in Fig. 3.4 on an enlarged scale.
tint(µs) 0.46 10 40 100 1000 2000 NETD (K) ≈44.00 2.00 0.50 0.20 0.02 0.01
Table 3.1: NETD values for differenttint.
3.1.4 Basic Image Processing
In this subsection, two basic image processing steps for the thermal images will be introduced. An advantage of the Thermosensorik CMT384 camera is that it records the raw count rates for each pixel and, thus, gives complete control over the image processing. One has to take into account that the dark level of each pixel in the array differs. This is no problem as long as the count rate has the same linear slope independent of the zero level. However, this calls for taking a reference picture to correct for the offset.
During each measurement campaign an image with long integration time, i.e., in the millisecond range, is taken. This gives information about the geometry. During a long tint a contrast between different emissivities of the materials can be detected, even if they have the same temperature. An example is given in Fig. 3.6(a-c). Figure 3.6(a) represents an image of the raw absolute pixel counts. Figure 3.6(b) is an image of an uniform area with constant temperature introduced between the object and the lens of the thermocamera, taken with the sametintas used for (a). These two images were taken with a temporal delay of just seconds. By taking the difference of (a) and (b) one gets image (c), where the discussed individual pixel-offset is corrected. In Fig. 3.6(c) one can clearly identify the geometric structure of the device, with the bond wires (marked by ’A’, one of the wires is additional indicated by a dashed line), the DL chip (’B’) and the sub-mount (’C’).
The flash of Planck’s radiation that is indicative for a COD event is recorded with much shortertintof usually 10µs. The detection of an emissivity contrast additional to the thermal flash is not possible under such conditions. However, the pixel-offset has to be taken into account, too. Therefore, an image of the unpowered device is taken immediately before the single pulse experiment. Both images were subtracted, Figs. 3.6(d) and (e). This leads to an image that reflects temperature differences, see Fig. 3.6(f). Notice, that also negative count rates around the zero line can appear, caused by this image subtraction.
In a last processing step, the two images, one giving the geometry [Fig. 3.6(e)], and one indicating the position of the COD event [Fig. 3.6(f)] can be composed, as long as the device is not moved between the image acquisitions. A threshold count-rate is chosen up to which the image (f) is set transparent, i.e., removing the parts of the image that have not changed its temperature during the single pulse.
Afterwards, it is overlaid to the background image. In order to distinguish between the data recorded with differenttint, the background is given in grayscale, while the thermal flash is represented in colorscale, see Fig. 3.6(g).
Figure 3.6: Basic image processing of thermal images. As an example the direct observation of a DLs’
front facet is shown. (a) Raw image of absolute count rates of each pixel. The image is taken withtint=1 ms. (b) Image of an uniform plane of constant temperature taken withtint=1 ms.
(c) Result of the subtraction of (b) from (a). The labeled objects are: bond wires (marked by ’A’, one of the wires is additional indicated by a dashed line), the DL chip (’B’) and the sub-mount (’C’). (d) Raw image taken during a COD event withtint=10µs. (e) Reference image taken of the un-powered device withtint=10µs. (f) Result of the subtraction of (e) from (d). (g) Composition of the images in (c) and (f). The area marked red is shown on an enlarged scale at the right side.
3.2 Streak Camera Based Setups
There are two reasons to use a streak camera for the present study:
• A temporal resolution of a line image in the picosecond time range, and
• continuous image-acquisition without any dead times.
The basic working principle of the camera is the following: Light enters the lateral entrance slit in front of the streak-tube. At the cathode, it is converted by the external photoelectric effect to a number of electrons according to light intensity. The electrons are accelerated in an electric field. They get deflected by a time-dependent electric field perpendicular to their flight direction and perpendicular to the lateral direction. This gives the temporal resolution while keeping the information about the lateral position at least in one dimension. Temporally dispensed this way, the electrons enter a multi-channel plate electron-multiplier. The resulting bunches of electrons hit a phosphor screen. The image visible on this screen is recorded by a CCD camera.
The experiments presented in this work used the Hamamatsu C1587 streak camera (with single sweep module M1953) to monitor the lateral intensity distribution of the lasing nearfield (NF) of the device. The DL was mounted on a heat sink and directly attached as close as possible to the used pulse driver Picolas LDP-V50-100 V3 to reduce the current rise time. In dependence on the width of the emitter, a microscope objective with appropriate magnification was chosen to image the NF onto the lateral entrance slit of the camera, see Fig. 3.7. In case of a multi-emitter device, e.g., an laser array, a simplified setup with a single lens was chosen. Two different streak-tubes were used, a S-1 tube which is infrared-enhanced (280µm-1550µm) and a S-20 one (200µm-900µm, higher sensitivity compared to S-1 in this range).
Figure 3.7: Scheme of the streak camera based setup.
